At least after 1950, when Alan Turing’s famous “Computing Machinery and Intelligence” paper was first published in the magazine The mind, Scientists interested in artificial intelligence have been fascinated by the notion of coding the mind. The mind, so the theory goes, is substrate independent, meaning that its processing capacity is not, by necessity, associated with the brain’s waitware. We can upload mindlessly on the computer or, conceptualize, create completely new people in the software world.
These are all familiar things. While we are yet to create or recreate a mind in software, out of the lowest-resolution abstraction that are modern neural networks, there is no shortage of computer scientists working on this effort at the moment.
The work being done by researchers at Tartu University in Estonia and Paris-Saikel University in France is less well known.
Instead of trying to recreate a presumption of brains in software, they have turned into a different problem: can you use an algorithm to generate genetic code for people who never existed The Can you apply the same generic adversarial network (GAN) technology that allows AI models to BigSelep to spit out real-time created images and instead use it to create fake DNA, which Is in the vein of Turing’s work, which is indistinguishable from it. Flesh and blood person?
Artificial genetic data
Flora J., a researcher specializing in machine learning and population genetics at the University of Paris-Saquel University in , said, “Creating artificial genetic data that is sufficiently realistic without replicating sequences is a very difficult problem. . ” “Genetic data is of high dimensions, and you just can’t know what’s important or not. That’s how we turned to cutting-edge technologies [being] Computer vision applies to the world of text, music or protein. These generator networks – GANs and [restricted Boltzmann machines] – are designed so that they can evolve progressively and automatically learn how to create an artificial genetic sequence. “
A GAN, a class of machine-learning frameworks coined by researcher (and current Apple employee) Ian Goodfellow, uses a combative, tug-of-war approach to improve its generic results. It consists of two neural networks: a “generator” and a “discriminator” that pass outputs between each other.
The generator’s job is to create something, it is an AI painting or a part of the code that represents an artificial genome in the form of people and zeros. Discriminating like the bot version of JK Simmons’ perfectionist music instructor in the film Sprain, Then criticizes its efforts and sends it back to the generator. The generator learns from this reaction, while the discriminator becomes ever better at guessing in the same way what the generator has created and what the actual article is. After all, the generator is so good at creating fake versions of whatever it is attempting to be fooled by the prudent. It is no longer able to distinguish real from fake.
“One of the main problems is the assessment of the quality of the artificial genome,” Burak Yelman, a Ph.D. A student at the University of Tartu’s Genomics told . “You can look at an image and decide if it looks real, but it is not possible for the genome. [The] Most of the analysis we did in our study was to see if the artificial genes we actually generated looked like real ones. “
Don’t worry though. Despite a growing mass of articles about highly suspected gene tampering designed to rewrite human code, the work is not about trying to “write” new parent humans who were created with the help of supercomputers can go.
“To be clear, the purpose of our work is to better understand and understand the existing genetic diversity of thousands or millions of people around the world, not to build artificial cells.” “Neural networks are trained on this existing diversity, so the generated genomic regions do not produce additional novel mutations that can easily disrupt the functionality of a sequence – and they include, untouched, segments that are conserved in human populations Huh.”
J stated that, on the whole genome scale, “it is difficult to say” whether a specific combination of millions of generated nucleotides could actually be “functional”. In other words, do not expect to compile and run this code, expecting a fully formed person (or their blueprint) at the other end. Instead, the objective is less sinister on the whole and, possibly, more useful.
All about data privacy
“Biobank has a huge amount of data and it keeps growing every day,” said Yelman. “However, genomic data is sensitive data and it may be difficult for researchers to access these biobanks due to ethical concerns. The main goal of our work is to create high-quality surrogates of existing genome banks and provide solutions to this access barrier within a secure ethical framework. It is important to note that our study was a first step: there is still work to be done.
Added J: “The idea behind our study is whether to start releasing artificial genes instead of real ones, providing useful information to the population genetics community. [Possible] Applications of artificial genomes can range from a better understanding of our genetic past to providing insights into medical genetics, including a wider range of diversity. “
In some ways, the work is reminiscent of the trend, seen a few years ago, in which GAN was used to create imaginary people, images of animals, and more as provided by the generative website ThisPersonDoesNameExist.com Was. Only this time, of course, it contains the actual genetic code rather than simple images.
A recent journal titled PLOS Genetics described a project titled “Creating artificial human genomes using man-made neural networks”.